Pull up the following revisions, requested by msaitoh in ticket #1795:

sys/dev/pci/if_wm.c				1.764-1.767 via patch
	sys/dev/pci/if_wmreg.h				1.128

- Workaround for some hypervisor environments. The environments
  cannot stop e1000 interrupt immediately.
- Rename nq_txdesc' member "nqrx_ctx" to "nqtx_ctx". No functional
  change.
- Add comment. Modify comment.
- KNF.
This commit is contained in:
martin 2023-01-23 14:01:25 +00:00
parent e21097e50c
commit d573b15002
2 changed files with 64 additions and 49 deletions

View File

@ -1,4 +1,4 @@
/* $NetBSD: if_wm.c,v 1.508.4.47 2022/09/08 10:29:36 martin Exp $ */ /* $NetBSD: if_wm.c,v 1.508.4.48 2023/01/23 14:01:25 martin Exp $ */
/* /*
* Copyright (c) 2001, 2002, 2003, 2004 Wasabi Systems, Inc. * Copyright (c) 2001, 2002, 2003, 2004 Wasabi Systems, Inc.
@ -82,7 +82,7 @@
*/ */
#include <sys/cdefs.h> #include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: if_wm.c,v 1.508.4.47 2022/09/08 10:29:36 martin Exp $"); __KERNEL_RCSID(0, "$NetBSD: if_wm.c,v 1.508.4.48 2023/01/23 14:01:25 martin Exp $");
#ifdef _KERNEL_OPT #ifdef _KERNEL_OPT
#include "opt_net_mpsafe.h" #include "opt_net_mpsafe.h"
@ -487,6 +487,7 @@ struct wm_queue {
char sysctlname[32]; /* Name for sysctl */ char sysctlname[32]; /* Name for sysctl */
bool wmq_txrx_use_workqueue; bool wmq_txrx_use_workqueue;
bool wmq_wq_enqueued;
struct work wmq_cookie; struct work wmq_cookie;
void *wmq_si; void *wmq_si;
krndsource_t rnd_source; /* random source */ krndsource_t rnd_source; /* random source */
@ -2263,7 +2264,7 @@ alloc_retry:
aprint_error_dev(sc->sc_dev, aprint_error_dev(sc->sc_dev,
"unable to find PCIX capability\n"); "unable to find PCIX capability\n");
else if (sc->sc_type != WM_T_82545_3 && else if (sc->sc_type != WM_T_82545_3 &&
sc->sc_type != WM_T_82546_3) { sc->sc_type != WM_T_82546_3) {
/* /*
* Work around a problem caused by the BIOS * Work around a problem caused by the BIOS
* setting the max memory read byte count * setting the max memory read byte count
@ -3633,7 +3634,7 @@ wm_watchdog_txq_locked(struct ifnet *ifp, struct wm_txqueue *txq,
ifp->if_oerrors++; ifp->if_oerrors++;
#ifdef WM_DEBUG #ifdef WM_DEBUG
for (i = txq->txq_sdirty; i != txq->txq_snext; for (i = txq->txq_sdirty; i != txq->txq_snext;
i = WM_NEXTTXS(txq, i)) { i = WM_NEXTTXS(txq, i)) {
txs = &txq->txq_soft[i]; txs = &txq->txq_soft[i];
printf("txs %d tx %d -> %d\n", printf("txs %d tx %d -> %d\n",
i, txs->txs_firstdesc, txs->txs_lastdesc); i, txs->txs_firstdesc, txs->txs_lastdesc);
@ -3794,9 +3795,9 @@ wm_tick(void *arg)
crcerrs + algnerrc + symerrc + rxerrc + sec + cexterr + rlec; crcerrs + algnerrc + symerrc + rxerrc + sec + cexterr + rlec;
/* /*
* WMREG_RNBC is incremented when there are no available buffers in host * WMREG_RNBC is incremented when there are no available buffers in
* memory. It does not mean the number of dropped packets, because an * host memory. It does not mean the number of dropped packets, because
* Ethernet controller can receive packets in such case if there is * an Ethernet controller can receive packets in such case if there is
* space in the phy's FIFO. * space in the phy's FIFO.
* *
* If you want to know the nubmer of WMREG_RMBC, you should use such as * If you want to know the nubmer of WMREG_RMBC, you should use such as
@ -4093,7 +4094,7 @@ wm_read_mac_addr(struct wm_softc *sc, uint8_t *enaddr)
return 0; return 0;
bad: bad:
return -1; return -1;
} }
@ -4378,11 +4379,11 @@ wm_set_filter(struct wm_softc *sc)
ifp->if_flags &= ~IFF_ALLMULTI; ifp->if_flags &= ~IFF_ALLMULTI;
goto setit; goto setit;
allmulti: allmulti:
ifp->if_flags |= IFF_ALLMULTI; ifp->if_flags |= IFF_ALLMULTI;
sc->sc_rctl |= RCTL_MPE; sc->sc_rctl |= RCTL_MPE;
setit: setit:
CSR_WRITE(sc, WMREG_RCTL, sc->sc_rctl); CSR_WRITE(sc, WMREG_RCTL, sc->sc_rctl);
} }
@ -4750,7 +4751,8 @@ wm_init_lcd_from_nvm(struct wm_softc *sc)
* LCD Write Enable bits are set in the NVM. When both NVM bits * LCD Write Enable bits are set in the NVM. When both NVM bits
* are cleared, SW will configure them instead. * are cleared, SW will configure them instead.
*/ */
DPRINTF(sc, WM_DEBUG_INIT, ("%s: %s: Configure SMBus and LED\n", DPRINTF(sc, WM_DEBUG_INIT,
("%s: %s: Configure SMBus and LED\n",
device_xname(sc->sc_dev), __func__)); device_xname(sc->sc_dev), __func__));
wm_write_smbus_addr(sc); wm_write_smbus_addr(sc);
@ -5056,15 +5058,15 @@ wm_initialize_hardware_bits(struct wm_softc *sc)
CSR_WRITE(sc, WMREG_TARC0, tarc0); CSR_WRITE(sc, WMREG_TARC0, tarc0);
switch (sc->sc_type) { switch (sc->sc_type) {
/*
* 8257[12] Errata No.52, 82573 Errata No.43 and some others.
* Avoid RSS Hash Value bug.
*/
case WM_T_82571: case WM_T_82571:
case WM_T_82572: case WM_T_82572:
case WM_T_82573: case WM_T_82573:
case WM_T_80003: case WM_T_80003:
case WM_T_ICH8: case WM_T_ICH8:
/*
* 8257[12] Errata No.52, 82573 Errata No.43 and some
* others to avoid RSS Hash Value bug.
*/
reg = CSR_READ(sc, WMREG_RFCTL); reg = CSR_READ(sc, WMREG_RFCTL);
reg |= WMREG_RFCTL_NEWIPV6EXDIS |WMREG_RFCTL_IPV6EXDIS; reg |= WMREG_RFCTL_NEWIPV6EXDIS |WMREG_RFCTL_IPV6EXDIS;
CSR_WRITE(sc, WMREG_RFCTL, reg); CSR_WRITE(sc, WMREG_RFCTL, reg);
@ -5468,6 +5470,11 @@ wm_reset(struct wm_softc *sc)
CSR_WRITE(sc, WMREG_CTRL, reg); CSR_WRITE(sc, WMREG_CTRL, reg);
/* Don't insert a completion barrier when reset */ /* Don't insert a completion barrier when reset */
delay(20*1000); delay(20*1000);
/*
* The EXTCNFCTR_MDIO_SW_OWNERSHIP bit is cleared by the reset,
* so don't use sc->phy.release(sc). Release sc_ich_phymtx
* only. See also wm_get_swflag_ich8lan().
*/
mutex_exit(sc->sc_ich_phymtx); mutex_exit(sc->sc_ich_phymtx);
break; break;
case WM_T_82580: case WM_T_82580:
@ -6091,7 +6098,7 @@ wm_setup_msix(struct wm_softc *sc)
kcpuset_destroy(affinity); kcpuset_destroy(affinity);
return 0; return 0;
fail: fail:
for (qidx = 0; qidx < txrx_established; qidx++) { for (qidx = 0; qidx < txrx_established; qidx++) {
struct wm_queue *wmq = &sc->sc_queue[qidx]; struct wm_queue *wmq = &sc->sc_queue[qidx];
pci_intr_disestablish(sc->sc_pc,sc->sc_ihs[wmq->wmq_intr_idx]); pci_intr_disestablish(sc->sc_pc,sc->sc_ihs[wmq->wmq_intr_idx]);
@ -6168,7 +6175,8 @@ wm_itrs_writereg(struct wm_softc *sc, struct wm_queue *wmq)
* So, overwrite counter field by software. * So, overwrite counter field by software.
*/ */
if (sc->sc_type == WM_T_82575) if (sc->sc_type == WM_T_82575)
eitr |= __SHIFTIN(wmq->wmq_itr, EITR_COUNTER_MASK_82575); eitr |= __SHIFTIN(wmq->wmq_itr,
EITR_COUNTER_MASK_82575);
else else
eitr |= EITR_CNT_INGR; eitr |= EITR_CNT_INGR;
@ -6269,7 +6277,8 @@ wm_init_sysctls(struct wm_softc *sc)
goto err; goto err;
rv = sysctl_createv(log, 0, &rnode, &cnode, CTLFLAG_READWRITE, rv = sysctl_createv(log, 0, &rnode, &cnode, CTLFLAG_READWRITE,
CTLTYPE_BOOL, "txrx_workqueue", SYSCTL_DESCR("Use workqueue for packet processing"), CTLTYPE_BOOL, "txrx_workqueue",
SYSCTL_DESCR("Use workqueue for packet processing"),
NULL, 0, &sc->sc_txrx_use_workqueue, 0, CTL_CREATE, CTL_EOL); NULL, 0, &sc->sc_txrx_use_workqueue, 0, CTL_CREATE, CTL_EOL);
if (rv != 0) if (rv != 0)
goto teardown; goto teardown;
@ -6971,7 +6980,7 @@ wm_init_locked(struct ifnet *ifp)
ifp->if_flags |= IFF_RUNNING; ifp->if_flags |= IFF_RUNNING;
ifp->if_flags &= ~IFF_OACTIVE; ifp->if_flags &= ~IFF_OACTIVE;
out: out:
sc->sc_if_flags = ifp->if_flags; sc->sc_if_flags = ifp->if_flags;
if (error) if (error)
log(LOG_ERR, "%s: interface not running\n", log(LOG_ERR, "%s: interface not running\n",
@ -7198,7 +7207,7 @@ wm_82547_txfifo_bugchk(struct wm_softc *sc, struct mbuf *m0)
return 1; return 1;
} }
send_packet: send_packet:
txq->txq_fifo_head += len; txq->txq_fifo_head += len;
if (txq->txq_fifo_head >= txq->txq_fifo_size) if (txq->txq_fifo_head >= txq->txq_fifo_size)
txq->txq_fifo_head -= txq->txq_fifo_size; txq->txq_fifo_head -= txq->txq_fifo_size;
@ -7263,14 +7272,14 @@ wm_alloc_tx_descs(struct wm_softc *sc, struct wm_txqueue *txq)
return 0; return 0;
fail_3: fail_3:
bus_dmamap_destroy(sc->sc_dmat, txq->txq_desc_dmamap); bus_dmamap_destroy(sc->sc_dmat, txq->txq_desc_dmamap);
fail_2: fail_2:
bus_dmamem_unmap(sc->sc_dmat, (void *)txq->txq_descs_u, bus_dmamem_unmap(sc->sc_dmat, (void *)txq->txq_descs_u,
WM_TXDESCS_SIZE(txq)); WM_TXDESCS_SIZE(txq));
fail_1: fail_1:
bus_dmamem_free(sc->sc_dmat, &txq->txq_desc_seg, txq->txq_desc_rseg); bus_dmamem_free(sc->sc_dmat, &txq->txq_desc_seg, txq->txq_desc_rseg);
fail_0: fail_0:
return error; return error;
} }
@ -7388,7 +7397,7 @@ wm_alloc_tx_buffer(struct wm_softc *sc, struct wm_txqueue *txq)
return 0; return 0;
fail: fail:
for (i = 0; i < WM_TXQUEUELEN(txq); i++) { for (i = 0; i < WM_TXQUEUELEN(txq); i++) {
if (txq->txq_soft[i].txs_dmamap != NULL) if (txq->txq_soft[i].txs_dmamap != NULL)
bus_dmamap_destroy(sc->sc_dmat, bus_dmamap_destroy(sc->sc_dmat,
@ -7579,7 +7588,7 @@ wm_alloc_txrx_queues(struct wm_softc *sc)
return 0; return 0;
fail_2: fail_2:
for (i = 0; i < rx_done; i++) { for (i = 0; i < rx_done; i++) {
struct wm_rxqueue *rxq = &sc->sc_queue[i].wmq_rxq; struct wm_rxqueue *rxq = &sc->sc_queue[i].wmq_rxq;
wm_free_rx_buffer(sc, rxq); wm_free_rx_buffer(sc, rxq);
@ -7587,7 +7596,7 @@ wm_alloc_txrx_queues(struct wm_softc *sc)
if (rxq->rxq_lock) if (rxq->rxq_lock)
mutex_obj_free(rxq->rxq_lock); mutex_obj_free(rxq->rxq_lock);
} }
fail_1: fail_1:
for (i = 0; i < tx_done; i++) { for (i = 0; i < tx_done; i++) {
struct wm_txqueue *txq = &sc->sc_queue[i].wmq_txq; struct wm_txqueue *txq = &sc->sc_queue[i].wmq_txq;
pcq_destroy(txq->txq_interq); pcq_destroy(txq->txq_interq);
@ -7599,7 +7608,7 @@ wm_alloc_txrx_queues(struct wm_softc *sc)
kmem_free(sc->sc_queue, kmem_free(sc->sc_queue,
sizeof(struct wm_queue) * sc->sc_nqueues); sizeof(struct wm_queue) * sc->sc_nqueues);
fail_0: fail_0:
return error; return error;
} }
@ -8806,12 +8815,12 @@ wm_nq_tx_offload(struct wm_softc *sc, struct wm_txqueue *txq,
* however it does not cause problems. * however it does not cause problems.
*/ */
/* Fill in the context descriptor. */ /* Fill in the context descriptor. */
txq->txq_nq_descs[txq->txq_next].nqrx_ctx.nqtxc_vl_len = txq->txq_nq_descs[txq->txq_next].nqtx_ctx.nqtxc_vl_len =
htole32(vl_len); htole32(vl_len);
txq->txq_nq_descs[txq->txq_next].nqrx_ctx.nqtxc_sn = 0; txq->txq_nq_descs[txq->txq_next].nqtx_ctx.nqtxc_sn = 0;
txq->txq_nq_descs[txq->txq_next].nqrx_ctx.nqtxc_cmd = txq->txq_nq_descs[txq->txq_next].nqtx_ctx.nqtxc_cmd =
htole32(cmdc); htole32(cmdc);
txq->txq_nq_descs[txq->txq_next].nqrx_ctx.nqtxc_mssidx = txq->txq_nq_descs[txq->txq_next].nqtx_ctx.nqtxc_mssidx =
htole32(mssidx); htole32(mssidx);
wm_cdtxsync(txq, txq->txq_next, 1, BUS_DMASYNC_PREWRITE); wm_cdtxsync(txq, txq->txq_next, 1, BUS_DMASYNC_PREWRITE);
DPRINTF(sc, WM_DEBUG_TX, DPRINTF(sc, WM_DEBUG_TX,
@ -8974,7 +8983,7 @@ wm_nq_send_common_locked(struct ifnet *ifp, struct wm_txqueue *txq,
DPRINTF(sc, WM_DEBUG_TX, DPRINTF(sc, WM_DEBUG_TX,
("%s: TX: have packet to transmit: %p\n", ("%s: TX: have packet to transmit: %p\n",
device_xname(sc->sc_dev), m0)); device_xname(sc->sc_dev), m0));
txs = &txq->txq_soft[txq->txq_snext]; txs = &txq->txq_soft[txq->txq_snext];
dmamap = txs->txs_dmamap; dmamap = txs->txs_dmamap;
@ -10075,9 +10084,13 @@ static inline void
wm_sched_handle_queue(struct wm_softc *sc, struct wm_queue *wmq) wm_sched_handle_queue(struct wm_softc *sc, struct wm_queue *wmq)
{ {
if (wmq->wmq_txrx_use_workqueue) if (wmq->wmq_txrx_use_workqueue) {
workqueue_enqueue(sc->sc_queue_wq, &wmq->wmq_cookie, curcpu()); if (!wmq->wmq_wq_enqueued) {
else wmq->wmq_wq_enqueued = true;
workqueue_enqueue(sc->sc_queue_wq, &wmq->wmq_cookie,
curcpu());
}
} else
softint_schedule(wmq->wmq_si); softint_schedule(wmq->wmq_si);
} }
@ -10376,8 +10389,10 @@ wm_handle_queue_work(struct work *wk, void *context)
struct wm_queue *wmq = container_of(wk, struct wm_queue, wmq_cookie); struct wm_queue *wmq = container_of(wk, struct wm_queue, wmq_cookie);
/* /*
* "enqueued flag" is not required here. * Some qemu environment workaround. They don't stop interrupt
* immediately.
*/ */
wmq->wmq_wq_enqueued = false;
wm_handle_queue(wmq); wm_handle_queue(wmq);
} }
@ -11824,7 +11839,7 @@ wm_disable_phy_wakeup_reg_access_bm(device_t dev, uint16_t *phy_regp)
*/ */
static int static int
wm_access_phy_wakeup_reg_bm(device_t dev, int offset, int16_t *val, int rd, wm_access_phy_wakeup_reg_bm(device_t dev, int offset, int16_t *val, int rd,
bool page_set) bool page_set)
{ {
struct wm_softc *sc = device_private(dev); struct wm_softc *sc = device_private(dev);
uint16_t regnum = BM_PHY_REG_NUM(offset); uint16_t regnum = BM_PHY_REG_NUM(offset);
@ -12191,8 +12206,8 @@ wm_gmii_statchg(struct ifnet *ifp)
CSR_WRITE(sc, WMREG_CTRL, sc->sc_ctrl); CSR_WRITE(sc, WMREG_CTRL, sc->sc_ctrl);
CSR_WRITE(sc, WMREG_TCTL, sc->sc_tctl); CSR_WRITE(sc, WMREG_TCTL, sc->sc_tctl);
CSR_WRITE(sc, (sc->sc_type < WM_T_82543) ? WMREG_OLD_FCRTL CSR_WRITE(sc, (sc->sc_type < WM_T_82543) ?
: WMREG_FCRTL, sc->sc_fcrtl); WMREG_OLD_FCRTL : WMREG_FCRTL, sc->sc_fcrtl);
if (sc->sc_type == WM_T_80003) { if (sc->sc_type == WM_T_80003) {
switch (IFM_SUBTYPE(mii->mii_media_active)) { switch (IFM_SUBTYPE(mii->mii_media_active)) {
case IFM_1000_T: case IFM_1000_T:
@ -14234,7 +14249,7 @@ wm_nvm_validate_checksum(struct wm_softc *sc)
DPRINTF(sc, WM_DEBUG_NVM, DPRINTF(sc, WM_DEBUG_NVM,
("%s: NVM need to be updated (%04x != %04x)\n", ("%s: NVM need to be updated (%04x != %04x)\n",
device_xname(sc->sc_dev), eeprom_data, device_xname(sc->sc_dev), eeprom_data,
valid_checksum)); valid_checksum));
} }
if ((sc->sc_debug & WM_DEBUG_NVM) != 0) { if ((sc->sc_debug & WM_DEBUG_NVM) != 0) {
@ -15221,8 +15236,8 @@ wm_init_phy_workarounds_pchlan(struct wm_softc *sc)
/* Acquire PHY semaphore */ /* Acquire PHY semaphore */
rv = sc->phy.acquire(sc); rv = sc->phy.acquire(sc);
if (rv != 0) { if (rv != 0) {
DPRINTF(sc, WM_DEBUG_INIT, ("%s: %s: failed\n", DPRINTF(sc, WM_DEBUG_INIT,
device_xname(sc->sc_dev), __func__)); ("%s: %s: failed\n", device_xname(sc->sc_dev), __func__));
return rv; return rv;
} }
@ -15468,8 +15483,8 @@ wm_ulp_disable(struct wm_softc *sc)
/* Acquire semaphore */ /* Acquire semaphore */
rv = sc->phy.acquire(sc); rv = sc->phy.acquire(sc);
if (rv != 0) { if (rv != 0) {
DPRINTF(sc, WM_DEBUG_INIT, ("%s: %s: failed\n", DPRINTF(sc, WM_DEBUG_INIT,
device_xname(sc->sc_dev), __func__)); ("%s: %s: failed\n", device_xname(sc->sc_dev), __func__));
return rv; return rv;
} }
@ -15482,7 +15497,7 @@ wm_ulp_disable(struct wm_softc *sc)
uint32_t reg2; uint32_t reg2;
aprint_debug_dev(sc->sc_dev, "%s: Force SMBus first.\n", aprint_debug_dev(sc->sc_dev, "%s: Force SMBus first.\n",
__func__); __func__);
reg2 = CSR_READ(sc, WMREG_CTRL_EXT); reg2 = CSR_READ(sc, WMREG_CTRL_EXT);
reg2 |= CTRL_EXT_FORCE_SMBUS; reg2 |= CTRL_EXT_FORCE_SMBUS;
CSR_WRITE(sc, WMREG_CTRL_EXT, reg2); CSR_WRITE(sc, WMREG_CTRL_EXT, reg2);
@ -15508,7 +15523,7 @@ wm_ulp_disable(struct wm_softc *sc)
wm_gmii_hv_writereg_locked(sc->sc_dev, 2, HV_PM_CTRL, phyreg); wm_gmii_hv_writereg_locked(sc->sc_dev, 2, HV_PM_CTRL, phyreg);
rv = wm_gmii_hv_readreg_locked(sc->sc_dev, 2, I218_ULP_CONFIG1, rv = wm_gmii_hv_readreg_locked(sc->sc_dev, 2, I218_ULP_CONFIG1,
&phyreg); &phyreg);
if (rv != 0) if (rv != 0)
goto release; goto release;
phyreg &= ~(I218_ULP_CONFIG1_IND phyreg &= ~(I218_ULP_CONFIG1_IND

View File

@ -1,4 +1,4 @@
/* $NetBSD: if_wmreg.h,v 1.98.6.15 2022/09/07 10:09:20 martin Exp $ */ /* $NetBSD: if_wmreg.h,v 1.98.6.16 2023/01/23 14:01:26 martin Exp $ */
/* /*
* Copyright (c) 2001 Wasabi Systems, Inc. * Copyright (c) 2001 Wasabi Systems, Inc.
@ -1703,7 +1703,7 @@ typedef union nq_txdesc {
uint32_t nqtxc_sn; uint32_t nqtxc_sn;
uint32_t nqtxc_cmd; uint32_t nqtxc_cmd;
uint32_t nqtxc_mssidx; uint32_t nqtxc_mssidx;
} nqrx_ctx; } nqtx_ctx;
} __packed nq_txdesc_t; } __packed nq_txdesc_t;